An article in Nature illustrates dangers in the training and output of AI systems, with uncertain human control over what the AI models learn
